Experimental Evaluation of Preprocessing Techniques in Constraint Satisfaction Problems

Abstract

This paper presents an evaluation of two orthogonal schemes for improving the efficiency of solving constraint satisfaction problems (CSPs). The first scheme involves a class of pre-processing techniques designed to make the representation of the CSP more explicit, including directional-arcconsistency, directional-path-consistency and adaptive-consistency. The second scheme aims at improving the order in which variables are chosen for evaluation during the search. In the first part of the experiment we tested the performance of backtracking (and its common enhancement -backjumping) with and without each of the preprocessings techniques above. The results show that directional arc-consistency, a scheme which embodies the simplest form of constraint recording, outperforms all other preprocessing techniques. The results of the second part of the experiment suggest that the best variable ordering is achieved by the fixed max-cardinality search order.